TFLiteConverter.from_冻结的图形无法加载冻结的图形：ValueE

Question

我使用tensorflow来定义一个图，将其量化并导出为冻结图。冻结的图形应转换为tensorflow光。我试着按照这里的说明：https://github.com/tensorflow/tensorflow/tree/r1.13/tensorflow/contrib/quantize

当我试图调用函数tf.lite.tflite转换器。从我得到错误：

tensorflow.python.framework.errors_impl.InvalidArgumentError: Input 0 of node conv2d/weights_quant/AssignMinLast was passed float from conv2d/weights_quant/min:0 incompatible with expected float_ref

这里有什么问题？ 我的代码：

    import tensorflow as tf
    from tensorflow.python.tools import freeze_graph

    with tf.Session() as sess:
        x = tf.placeholder(tf.float32, [None, 27648], name="x")  # 128*72*3
        y_ = tf.placeholder(tf.float32, [None, 72, 128, 1], name="y_")  # 128*72*1

        # 'reshape':
        input_layer = tf.reshape(x, [-1, 72, 128, 3])

        # Convolutional Layer
        conv1 = tf.layers.conv2d(
            inputs=input_layer,
            filters=32,
            kernel_size=[3, 3],
            padding="same",
            activation=tf.nn.relu,
            )
        y_out = tf.math.multiply(
            conv1,
            1.,
            name="y_out"
        )

        loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_out))

        tf.initialize_all_variables().run()

        # Build training model
        g = tf.get_default_graph()
        tf.contrib.quantize.create_training_graph(input_graph=g,
                                                  quant_delay=2000000)
        tf.initialize_all_variables().run()

        # Build eval model
        lossEval = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_out))
        # Call the eval rewrite which rewrites the graph in-place with
        # FakeQuantization nodes and fold batchnorm for eval.
        g = tf.get_default_graph()

        tf.contrib.quantize.create_eval_graph(input_graph=g)

        # Call backward pass optimizer as usual.
        learning_rate = 5e-5
        optimizer = tf.train.GradientDescentOptimizer(learning_rate)
        optimizer.minimize(loss)

        writer = tf.summary.FileWriter('./graphs', sess.graph)

        #todo: train

        export_path = './export/'
        print('Exporting trained model to', export_path)

        with open(export_path + "qtrained.pbtxt", 'w') as f:
            f.write(str(g.as_graph_def()))
        saver = tf.train.Saver()
        saver.save(sess, export_path+"qtrained.ckpt")

        for n in tf.get_default_graph().as_graph_def().node:
            print(n.name)

        freeze_graph.freeze_graph(input_graph = export_path + "qtrained.pbtxt",
                     input_saver = "",
                     input_binary = False,
                     input_checkpoint = export_path + "./qtrained.ckpt",
                     output_node_names = "y_out",
                     restore_op_name = "save/restore_all",
                     filename_tensor_name = "save/Const:0",
                     output_graph = './frozen/frozen_model.pb',
                     clear_devices = False,
                     initializer_nodes = "")


        converter = tf.lite.TFLiteConverter.from_frozen_graph('./frozen/frozen_model.pb', ['x'], ['y_out'])
        converter.inference_type = tf.lite.constants.QUANTIZED_UINT8
        input_arrays = converter.get_input_arrays()
        converter.quantized_input_stats = {input_arrays[0]: (0., 1.)}  # mean, std_dev
        tflite_model = converter.convert()
        open("new_converted_model.tflite", "wb").write(tflite_model)
        print("converting done")